new group#31682
Merged
seiriosPlus merged 3 commits intoPaddlePaddle:developfrom Apr 1, 2021
kuizhiqing:develop
Merged
Conversation
|
|
|
Thanks for your contribution! |
|
✅ This PR's description meets the template requirements! |
qili93
requested changes
Mar 23, 2021
gongweibao
requested changes
Mar 23, 2021
Contributor
gongweibao
left a comment
gongweibao
left a comment
There was a problem hiding this comment.
Please finetune you title and repair your CI first!
sandyhouse
reviewed
Mar 24, 2021
gongweibao
requested changes
Mar 29, 2021
| return gp | ||
|
|
||
|
|
||
| def wait(tensor, group=None, use_calc_stream=True): |
Contributor
There was a problem hiding this comment.
有个疑问
- use_calc_stream这个参数是用来做什么的？设置为True或者False的时候，分别有什么作用？
- 未来是否可能会增加对其他stream的控制？
Member
Author
There was a problem hiding this comment.
- paddle 对 gpu 的 stream 进行的逻辑抽象，calculation stream 和 communication stream，对应不同的通道；
- 不会，除了这层抽象外，还有多个 comm stream，用 id 做区分，和 group 绑定。
| attrs={'ring_id': ring_id}, ) | ||
|
|
||
|
|
||
| def broadcast(tensor, src, group=None, use_calc_stream=True): |
Contributor
There was a problem hiding this comment.
group=0改为group=None，对兼容性是否会有影响？
比如，是否有代码设定group=1的情况？
Member
Author
There was a problem hiding this comment.
已经和同事确认过 group=None 不会有影响，在本次 pr 前无法创建 group，所以 group=1的情况不存在，另外 group=0 显示调用的情况在代码中已经排除。
TCChenlong
reviewed
Apr 1, 2021
| def new_group(ranks=None, backend=None): | ||
| """ | ||
|
|
||
| Creates a new distributed comminication group. |
| backend (str): The backend used to create group, only nccl is supported now. | ||
|
|
||
| Returns: | ||
| Group: The group instance. Nerver return None. |
| import paddle | ||
|
|
||
| paddle.distributed.init_parallel_env() | ||
| tindata = np.random.random([10, 1000]).astype('float32') |
Contributor
There was a problem hiding this comment.
用paddle.random 新建Tensor，就不用使用 numpy了
| Examples: | ||
| .. code-block:: python | ||
|
|
||
| import numpy as np |
| import paddle | ||
|
|
||
| paddle.distributed.init_parallel_env() | ||
| tindata = np.random.random([10, 1000]).astype('float32') |
jzhang533
reviewed
Apr 1, 2021
|
|
||
| paddle.distributed.init_parallel_env() | ||
| tindata = np.random.random([10, 1000]).astype('float32') | ||
| tindata = paddle.to_tensor(tindata) |
|
|
||
| Args: | ||
| ranks (list): The global ranks of group members, list as sorted. | ||
| backend (str): The backend used to create group, only nccl is supported now. |
Contributor
There was a problem hiding this comment.
backend设定默认值为None时，现在的行为是直接设置为用nccl，未来有计划改这个默认行为吗？
| place = core.CUDAPlace(genv.device_id) | ||
| core.NCCLParallelContext(strategy, place).init_with_ring_id(ring_id) | ||
| else: | ||
| assert False |
| Creates a new distributed comminication group. | ||
|
|
||
| Args: | ||
| ranks (list): The global ranks of group members, list as sorted. |
Contributor
There was a problem hiding this comment.
没太懂list as sorted，是啥意思，是对ranks里的值的序有要求？
| Examples: | ||
| .. code-block:: python | ||
|
|
||
| import numpy as np |
|
|
||
| paddle.distributed.init_parallel_env() | ||
| tindata = np.random.random([10, 1000]).astype('float32') | ||
| tindata = paddle.to_tensor(tindata) |
| @@ -163,7 +371,9 @@ def all_reduce(tensor, op=ReduceOp.SUM, group=0): | |||
| tensor (Tensor): The input Tensor. It also works as the output Tensor. Its data type | |||
| should be float16, float32, float64, int32 or int64. | |||
| op (ReduceOp.SUM|ReduceOp.MAX|ReduceOp.Min|ReduceOp.PROD): Optional. The operation used. | |||
| @@ -238,7 +454,9 @@ def reduce(tensor, dst, op=ReduceOp.SUM, group=0): | |||
| should be float16, float32, float64, int32 or int64. | |||
| dst (int): The destination rank id. | |||
| op (ReduceOp.SUM|ReduceOp.MAX|ReduceOp.Min|ReduceOp.PROD): Optional. The operation used. | |||
| @@ -394,7 +626,9 @@ def scatter(tensor, tensor_list=None, src=0, group=0): | |||
| tensor_list (list): A list of Tensors to scatter. Every element in the list must be a Tensor whose data type | |||
| should be float16, float32, float64, int32 or int64. | |||
| src (int): The source rank id. | |||
| @@ -394,7 +626,9 @@ def scatter(tensor, tensor_list=None, src=0, group=0): | |||
| tensor_list (list): A list of Tensors to scatter. Every element in the list must be a Tensor whose data type | |||
TCChenlong
approved these changes
Apr 1, 2021
Contributor
TCChenlong
left a comment
TCChenlong
left a comment
There was a problem hiding this comment.
LGTM
TODO：fix docs bug
jzhang533
approved these changes
Apr 1, 2021
Contributor
jzhang533
left a comment
jzhang533
left a comment
There was a problem hiding this comment.
lgtm
will fix doc problem in following pr
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
10 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
APIs
Describe
unitest coverd by